AVs are safe.
Now we need safety beliefs.

Let's start with what the road data says. Across more than 170 million fully driverless miles, Waymo reports 92% fewer crashes involving serious injury or worse than matched human benchmarks — 0.02 such crashes per million miles versus 0.22 for human drivers — along with large reductions in airbag deployments and pedestrian injuries. A peer-reviewed analysis of the first 56.7 million of those miles found the same pattern. Reasonable people can debate edge cases and reporting methods, but the central finding here is not ambiguous: today's leading AVs crash less, and less severely, than we humans do.

Unfortunately, this remarkable safety record doesn't actually determine adoption by itself. We need prospective riders to believe this, and then make the decision to ride. But almost nobody knows this. In our 2025 national poll, Americans split nearly evenly on whether traffic accidents are more likely with AVs or with human drivers — 28% said human drivers, 31% said AVs, and four in ten either saw no difference or couldn't say. Only about one in four Americans hold the belief that the crash data supports.

Everything depends on one belief.

Why does this single belief matter so much? Because support for AVs isn't really about AVs. It's about what people think AVs will do. In our data, support for allowing robotaxi service locally correlates with the belief that AVs are safer than human drivers at r = 0.443 — in survey research, that's the two moving together almost in lockstep. Overall, Americans oppose local AV deployment 44% to 29%. But if we zoom in to just people who believe accidents are much more likely with human drivers, that flips to roughly 60% support and 20% opposition (still some undecideds). It's the same technology and the same question, but having a different core belief changes everything.

Even the demographic gaps resolve into this one variable. Men support AVs far more than women — but men are also far more likely to believe AVs are safer. So even apparent differences across potential market segments might be largely explained simply by differences in these core beliefs about safety.

The chain from belief to behavior

The load-bearing function of beliefs is well-documented in the research on persuasion and behavior-change. For example, the Theory of Reasoned Action and its successor, the Theory of Planned Behavior, have organized decades of research around a simple causal chain: behavior follows intention, intention follows attitude, and attitude is built from underlying beliefs about that behavior and its effects. At the core, most people ride in an AV taxi because they believe the ride will be safe, affordable, uncomplicated, and successful — and those beliefs, weighted by how much you care about each outcome, become your attitude, which becomes your intention, which eventually can convert into behavior when you have an opportunity.

The practical implication is that the industry's persuasion problem lives at the top of the chain, not the bottom. Advertising the experience — the smooth ride, the privacy, the novelty — targets attitudes directly while leaving the load-bearing belief untouched. As long as someone believes AVs are more dangerous than human drivers, every downstream appeal is pushing against a locked door. But if we can change the safety belief, then the rest of the chain starts to move on its own. In fact, our own data shows majority support already exists among believers, no further persuasion required.

What actually moves the belief?

Three levers of belief change show up consistently in the data. First, knowledge of current reality: half of Americans didn't even know AV taxis were already operating in several U.S. cities. That awareness (or lack of it) was one of the strongest correlates of opposition we measured. That is, people who thought AVs were "far from real-world use" opposed local deployment by roughly two to one. Second, direct and social exposure: in cities where robotaxis operate, favorability has climbed dramatically as rides became ordinary. Third, trusted messengers: no AV developer yet earns majority trust, and the brands most visible in deployment rank near the bottom — which means the safety evidence needs carriers the public already believes.

The technology will keep improving. But adoption won't necessarily follow in line with the engineering. For AVs — as for most innovations we study — the decisive competition isn't on the road. It's in the mind.